A Clinical Food Image Dataset for AI-Based Dietary Assessment in Hospital Care

Reliable assessment of food intake is a challenge in hospital nutrition care, where manual recording and visual estimation can limit accuracy, reproducibility, and scalability. Image-based methods may support objective dietary monitoring, but their development requires datasets with measured portion mass and nutritional annotations collected under clinically relevant conditions. In this work, we present ClinicPlate, a food image dataset created for quantitative food portion and macronutrient estimation in clinical environment. The dataset contains 9,868 images of individual food portions from 7 hospital food categories: boiled fish, carrots in oil, cooked ham, egg omelette, steamed fennel, white rice, and zucchini. For each category, 10 portion levels were prepared, weighed with 1g precision, and photographed on a standardized white plate and tray setup in a controlled hospital environment. Each image is associated with food category, portion level, measured weight, and macronutrients information. The ClinicPlate dataset can be used as a valuable resource for the development, training, and validation of artificial intelligence and computer vision methods aimed at supporting dietary monitoring in hospital settings. Given the absence of patients' personal information in the dataset, its use may also extend to other settings requiring quantitative food intake assessment.